{"id":"a97a55af-bba7-469a-9cfd-19f0247f1ee4","arxiv_id":"1908.04838","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"New DNS up to 2048^3 grid points reportedly confirm that perturbations in fully developed turbulence amplify as e^{c sqrt(Re) sqrt(t)}, which is faster than exponential, and saturate quickly.","lead":"This paper uses high-resolution computer simulations of turbulent fluids to test the idea that tiny perturbations grow faster than exponentially, at a rate that also grows with the square root of the Reynolds number. The authors say the simulations confirm this 'superfast' growth and argue that it explains how turbulence develops and keeps itself going.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reynolds-number scaling is not settled: the paper's own Figure 3 includes a comparable Re^0.38 fit, and no residuals, error bars, or model-selection statistics are given.","rationale":"The reader's weakest_assumption identifies the fitting window and the unverified constant relation as the soft spot. I agree, though I would sharpen the concern further: the decisive weakness is that the paper's own Figure 3 offers a competing Re^0.38 fit without any statistical discrimination, so the claimed √Re verification is not established by the presented evidence. This is internal to the paper, not a matter of disagreement with external consensus. The underlying DNS work is substantial, with resolutions up to 2048^3 and Re up to 6210, and the prediction is specific and falsifiable, which deserves credit. But the inferential step from noisy, short-window curves to a confirmed universal law is not quantitatively supported. The correct disposition remains CONDITIONAL: the hypothesis is plausible and testable, but this paper does not settle it. Therefore the reader's verdict is unchanged.","tokens_in":5454,"tokens_out":3089,"duration_ms":33636,"concrete_test":"Re-analyze the same DNS runs (or repeat them in an existing forced isotropic code) as follows: for each Re, compute the linearized amplitude δu(t) before saturation, and fit ln δ = a√t + b t over the claimed plateau window with bootstrap error bars; compare this model against ln δ = λ t and ln δ = a t^p using AIC or residual sums. Then, for the Reynolds scaling, fit ln Δ(0.3T0) against both √Re and Re^0.38 as separate one-parameter fits, and report the residual norms and parameter uncertainties. Also record the time at which linear and nonlinear amplitudes diverge for each Re; if any high-Re case has already saturated before 0.3T0, Figure 3's interpretation is invalid.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central confirmation claim rests on two scalings: ln(Δ/Δ0) ~ √t at fixed Re and ln(Δ/Δ0) ~ √Re at fixed t. The time scaling is inferred from visual plateaus in [ln(Δ/Δ0)] t^{-1/2} over t in (0, ~1), with the t→0 limit excluded as a numerical artifact. Over such a short window, any growth law of the form ln A ~ a t^p with p near 1/2, or a smooth crossover between regimes, can produce nearly flat segments; no quantitative goodness-of-fit or comparison with e^{λt} or t^p alternatives is reported. The Reynolds scaling is even less secure: Figure 3 plots a single time, t = 0.3 T0, and displays two fits, one ∼√Re and one ∼Re^0.38, with no residual analysis or parameter uncertainties. If Re^0.38 fits comparably, the data do not discriminate the predicted √Re dependence. Additionally, the measured quantity is the nonlinear amplitude Δ(t), while prediction (1) is for the linear amplitude δu(t); the paper asserts without quantitative evidence that t = 0.3 T0 lies in the pre-saturation linear regime for all Re shown. If high-Re runs have already partially saturated by that time, the observed amplitude reflects saturation rather than superfast linear growth, so the plateau would not confirm e^{σ√Re√t + σ1 t}.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports a numerical verification of the authors' earlier prediction that the maximal amplification of perturbations in fully developed homogeneous isotropic turbulence grows as e^{σ√Re√t + σ1 t}, i.e. faster than exponential, with σ1 = (√e/2)σ. Direct numerical simulations of forced Navier-Stokes turbulence at resolutions up to 2048^3 and Reynolds numbers up to 6210 are used. The evidence is presented as: (i) in Figure 2, plots of [ln(Δ(t)/Δ(0))] t^{-1/2} versus t for four Reynolds numbers, whose horizontal segments are interpreted as demonstrating the √t growth in the exponent; and (ii) in Figure 3, a plot of ln(Δ(0.3T0)/Δ(0)) versus Re with a fit to ~√Re, alongside a competing Re^0.38 fit. The paper concludes that superfast amplification and superfast nonlinear saturation of ever-present perturbations constitute the mechanism for generation, development, and persistence of turbulence.","tokens_in":5773,"tokens_out":2725,"duration_ms":29052,"significance":"If the predicted law (1) were convincingly confirmed and the constants σ and σ1 quantitatively matched, this would be a significant result: it would place turbulence in a regime more violent than low-dimensional chaos, contradicting Ruelle's exponential Lyapunov scaling, and would offer a mechanistic explanation for turbulence persistence. The computational effort is a genuine strength: DNS up to 2048^3 with Re up to 6210 is a substantial campaign, and the manuscript carefully distinguishes nonlinear amplitude Δ(t) from linear amplitude δu(t), an important conceptual point. However, the verification as presented is qualitative rather than quantitative. The confirmation claims rest on visual plateaus in Figure 2 over less than one large-eddy turnover time, and on a single-time Reynolds-number plot in Figure 3 that includes a visually comparable Re^0.38 fit, with no error bars, residuals, or model-selection statistics. Because the central claim is precisely a specific functional form with specific constants, the lack of quantitative fitting is a load-bearing gap. The result may well be correct, but the evidence in this manuscript does not yet establish it.","major_comments":[{"comment":"The evidence for the e^{c√t} behavior is only the visual flatness of [ln(Δ(t)/Δ(0))] t^{-1/2} over roughly 0 ≤ t ≤ 1 (about half a large-eddy turnover time, since T0 ≈ 2). No error bars, no goodness-of-fit measure, and no comparison against alternative growth laws such as e^{λt} or t^p with p near 1/2 are provided. Over such a short window, many functional forms produce nearly flat segments, especially after excluding the t→0 region. The paper should report quantitative fits to log-amplitude versus t for each Re, with parameter uncertainties and a model-selection statistic (e.g., AIC or a chi-square ratio) that discriminates √t from exponential or other powers. Without this, the plateau observation does not by itself confirm the √t scaling.","section":"Figure 2 and surrounding text"},{"comment":"The Reynolds-number scaling is examined at a single time t = 0.3T0, and the figure itself displays two fits, one ~√Re and one ~Re^0.38. The text asserts that the data fit well with e^{c√Re}, but no residuals, parameter uncertainties, or statistical comparison between the two fits are given. Visually, the two curves are close over the plotted range, so the data do not discriminate the predicted √Re dependence from a slower power law. A quantitative fit with confidence intervals, or a plot of ln(amplification) versus √Re with the Re^0.38 curve and residuals, is needed to support the central claim.","section":"Figure 3 and surrounding text"},{"comment":"Prediction (1) is for the linear amplification δu(t), while the verification plots use the nonlinear amplitude Δ(t). The paper states that during the early stage the linear amplification is a good approximation of the nonlinear one, but no quantitative evidence is given that t = 0.3T0 is within the pre-saturation linear regime for all Reynolds numbers shown. Since saturation occurs sooner at higher Re (as the text itself notes), the high-Re points in Figure 3 may already be in the nonlinear saturation regime, in which case the observed amplitude reflects saturation rather than superfast linear growth. The authors should show, for each Re, the divergence time between δu(t) and Δu(t) and confirm that the fitting window lies before it, or present linear-amplification data directly.","section":"Equations (1), (5)-(6) and Figures 2-3"},{"comment":"The prediction (1), including the relation σ1 = (√e/2)σ, is imported from the authors' prior work [19-22] and is not rederived in this manuscript. The numerical tests, however, use a generic constant c and never compare the fitted coefficient with the predicted σ or σ1. Consequently, the verification is only of the functional form with free parameters, not of the quantitative prediction. To claim confirmation, the manuscript should either derive the constants within the present framework or explicitly test whether the measured prefactor matches the predicted value within uncertainties. As it stands, the functional-form test alone is a weaker statement than the paper's conclusion.","section":"Equation (1) and the constants σ, σ1"}],"minor_comments":[{"comment":"There are typographical errors: 'meterology' should be 'meteorology' and 'undertaining' should be 'understanding'.","section":"Introduction (paragraph 1)"},{"comment":"Reference [23] is cited as an arXiv preprint (arXiv:1702.02993, 2018); if it has been published in a journal, the published version should be cited, and if not, the manuscript should indicate that the verification builds on an unpublished preprint.","section":"References"},{"comment":"The phrase 'Except in t→0+ limit' is vague: the excluded region is not precisely defined. For reproducibility, the authors should specify the time interval over which the fits are performed and how the initial transient is excluded.","section":"Figure 2 caption and text"},{"comment":"The figure uses both a dashed curve and a red/grey curve but the caption does not explicitly state which is which; adding labels directly in the figure or a clear legend would improve clarity.","section":"Figure 3"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's central claim is striking, but the verification is not yet at the standard expected for a claim that turbulence is generated by superfast amplification. The authors should be encouraged to provide a quantitative statistical analysis of the fits, including error bars and model comparison, and to directly compare linear and nonlinear amplification in the fitting window. The reliance on the authors' own prior work for the constants is also a concern; a rederivation or a test of the prefactor would strengthen the paper considerably. The paper is within the scope of physics.flu-dyn, but the current evidence is insufficient for acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nShort version: this is a higher-resolution follow-up to the authors' own earlier prediction, and the numerics are real work—but the paper's conclusion that the data confirm the predicted e^{σ√Re√t+σ1t} law is not actually supported by the analysis as written.\n\nWhat's new: they run DNS up to 2048^3 and Re=6210 in forced homogeneous isotropic turbulence, reaching the fully developed regime, and show that perturbation amplification curves, plotted as [ln(Δ/Δ0)]/√t, flatten into plateaus over t ≈ 0.2–0.8. That is a reasonable visual check. They also show at higher Re that the nonlinear amplification saturates faster. This genuinely extends their earlier low-resolution work and deserves credit.\n\nWhere it strains: the evidence is qualitative. The plateau in Figure 2 sits over a short time window, and the t→0 limit is discarded by fiat. An exponential e^{λt} would also produce a falling curve in this plot, so without fitting an alternative law the claim that √t wins is not tested. Figure 3 is the bigger problem: they plot a single time, t=0.3T0, and include both a √Re fit and a Re^0.38 fit. With no residuals or error bars, the data do not discriminate between them. Their own plot shows the two curves are comparable.\n\nAnother soft spot: prediction (1) is for the linear amplification δu, but they measure the nonlinear Δ and assert, without quantitative evidence, that t=0.3T0 lies in the pre-saturation linear regime for every Re shown. High-Re runs likely saturate earlier, and once saturated the observed amplitude says nothing about superfast linear growth. They also test only e^{c√t} and e^{c√Re} with generic constants c, never checking whether the fitted c matches the predicted σ or σ1, including σ1=(√e/2)σ. So the verification is of a functional form only, not the quantitative prediction.\n\nWho is this for? Specialists in turbulent perturbation growth and the chaos-versus-turbulence debate. They get a clearly written short report with a serious DNS run. But the analysis needs a revision pass: error bars, a proper model comparison (exponential vs √t), a check on saturation time per Re, and ideally a direct linearized simulation to compare δu with Δ. As it stands, building a mechanism-of-turbulence claim on this evidence is too strong.\n\nVerdict: worth a serious referee, but conditional acceptance after revision, not as-is.\n\nRegards.","headline":"Real DNS effort and a legitimate visual check of the superfast amplification idea, but the confirmation claim outruns the evidence: no error bars, an unresolved Re^0.38 vs √Re ambiguity in the paper's own Figure 3, and no test of the predicted constants.","tokens_in":6290,"tokens_out":2379,"would_cite":false,"duration_ms":23251,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["47.27.Gs","05.45.-a","47.27.ek"],"model":"deepseek-v4-flash","headline":"This paper reports that fully developed turbulence amplifies initial perturbations according to $e^{\\sigma\\sqrt{Re}\\sqrt{t}+\\sigma_1 t}$ — faster than exponential — and that direct numerical simulations up to Reynolds number 6210 confirm…","keywords":["turbulence mechanism","superfast amplification","perturbation growth","Navier-Stokes equations","direct numerical simulation","Lyapunov exponent","nonlinear saturation","homogeneous isotropic turbulence"],"falsifier":"Run the same linear perturbation setup at Reynolds number around 6000 and fit $\\ln(\\delta u(t)/\\delta u(0))/\\sqrt{t}$ over $0.1T_0$ to $0.5T_0$; if the plateau tilts or drifts systematically with the fit window, or if the amplification at fixed $t=0.3T_0$ scales as $Re^{0.38}$ rather than $e^{c\\sqrt{Re}}$, the predicted superfast law is not confirmed.","tokens_in":5264,"feed_emoji":"🌪️","tokens_out":5765,"duration_ms":52567,"temperature":0.7,"pith_summary":"The paper proposes a mechanism for fully developed turbulence: ever-present small perturbations are amplified superfast, growing as $e^{\\sigma\\sqrt{Re}\\sqrt{t}+\\sigma_1 t}$ in the early stage, and this growth drives a superfast nonlinear saturation. Such growth is faster than the exponential growth that characterizes deterministic chaos, matching the intuition that turbulence is more violent than chaos. The authors run direct numerical simulations of the three-dimensional Navier-Stokes equations in homogeneous isotropic turbulence, with resolution up to $2048^3$ and Reynolds numbers up to 6210, and report that the measured amplification follows the predicted $\\sqrt{t}$ dependence in time and $\\sqrt{Re}$ dependence in Reynolds number until nonlinear saturation. If correct, the result identifies the generation, development, and persistence of turbulence with this superfast amplification rather than with classical Lyapunov chaos.","feed_headline":"Turbulence amplifies perturbations faster than exponential","feed_subtitle":"New simulations to Reynolds 6210 back superfast amplification as the engine of turbulence.","key_machinery":"The load-bearing object is the explicit growth law $e^{\\sigma\\sqrt{Re}\\sqrt{t}+\\sigma_1 t}$, a prediction the authors derive from rigorous analysis of the Navier-Stokes equations and previously tested at low resolution. The verification mechanism is the Eulerian perturbation-tracking setup: a statistically steady turbulent field is duplicated, one copy has the forcing skipped for one time step, and the linear and nonlinear perturbation evolutions are followed by subtracting fields. Matching the measured amplification curves to the predicted $\\sqrt{t}$ plateaus and the $\\sqrt{Re}$ scaling at fixed $0.3T_0$ is what carries the confirmation. The relation $\\sigma_1=(\\sqrt{e}/2)\\sigma$ fixes the relative size of the exponential correction term.","core_discovery":"On the paper's own terms, the central discovery is that perturbation amplification in fully developed homogeneous isotropic turbulence is governed by $e^{\\sigma\\sqrt{Re}\\sqrt{t}+\\sigma_1 t}$, with $\\sigma_1=(\\sqrt{e}/2)\\sigma$, so the early-time amplification is much faster than exponential. The numerical evidence is the key claim: for Reynolds numbers 130, 805, 1450, and 2520, plots of $[\\ln(\\Delta(t)/\\Delta(0))]t^{-1/2}$ show horizontal plateaus over an extended interval, indicating $e^{c\\sqrt{t}}$ growth, and the amplification at $0.3T_0$ grows with Reynolds number roughly as $e^{c\\sqrt{Re}}$. The same superfast growth is argued to lead naturally to superfast nonlinear saturation, so turbulence is generated, developed, and maintained by the relentless amplification of perturbations that already exist in the flow.","pith_inferences":["A natural testable extension is to measure the finite-time Lyapunov exponent $\\Lambda(t)=\\ln(\\delta u(t)/\\delta u(0))/t$; the superfast law predicts $\\Lambda(t)\\sim \\sigma\\sqrt{Re}/\\sqrt{t}$ at early times, so the standard long-time Lyapunov exponent may not exist or may be dominated by the $\\sigma_1$ term.","The predicted universality of $\\sigma$ and $\\sigma_1$ could be checked by varying the forcing band $k_f$ and the initial perturbation band; the paper fixes $k_f=2.5$ and injects the perturbation in that band, so it does not test whether the constants are truly universal.","If the $\\sqrt{Re}$ scaling holds up to very high Reynolds numbers, then Kolmogorov-based dimensional estimates of Lyapunov exponents need revision, because the relevant time scale is the combination $1/(\\sigma^2 Re)$ from the superfast term rather than the Kolmogorov time."],"forward_implications":["If the growth law is correct, fully developed turbulence amplifies perturbations several times faster than low-dimensional chaos, since chaos gives at most exponential growth.","Predictability windows shrink superfast: errors grow like $e^{c\\sqrt{t}}$, so the time to reach a given error level scales roughly as $(\\text{error}/\\sigma\\sqrt{Re})^2$, much shorter than in exponential growth.","Nonlinear saturation time drops as Reynolds number rises, so higher-Re flows saturate perturbations sooner, consistent with the observed violence of developed turbulence.","The authors state the theory should guide turbulence engineering and ensemble weather forecasting, where initial-condition uncertainty must be tracked."],"supporting_citations":[{"why":"Supplies the baseline exponential prediction with maximal Lyapunov exponent proportional to $\\sqrt{Re}$ that the superfast law extends.","marker":"[12]"},{"why":"Establishes boundedness of the nonlinear perturbation amplification, supporting the distinction between linear and nonlinear growth.","marker":"[18]"},{"why":"One of the earlier analytic derivations of the superfast growth law from the Navier-Stokes equations.","marker":"[19]"},{"why":"Further analytic support for the predicted $e^{\\sigma\\sqrt{Re}\\sqrt{t}+\\sigma_1 t}$ growth law.","marker":"[20]"},{"why":"Rigorous input on the Navier-Stokes dynamics used in the analytic derivation of the prediction.","marker":"[21]"},{"why":"Continues the analytic derivation and fixes the constant relation $\\sigma_1=(\\sqrt{e}/2)\\sigma$ used in the prediction.","marker":"[22]"},{"why":"Earlier low-resolution numerical verification that first tested the predicted growth law before the present large-scale simulations.","marker":"[23]"},{"why":"Provides the de-aliased pseudo-spectral code used for the direct numerical simulations.","marker":"[24]"},{"why":"Supplies the external forcing scheme that drives the statistically steady homogeneous isotropic turbulence.","marker":"[25]"}],"fun_headline_variants":["Superfast growth: perturbations outpace exponential","Turbulence engine: superfast amplification, saturation","Perturbations grow faster than exponential in turbulence","Superfast amplification: turbulence's self-sustaining secret","Beyond exponential: how turbulence amplifies perturbations"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire confirmation rests on the assumption that the time window over which the fits are made (roughly up to $0.3$–$0.5$ large-eddy turnover times) lies entirely inside the superfast $e^{c\\sqrt{t}}$ regime, and that the constants $\\sigma$ and $\\sigma_1$, including $\\sigma_1=(\\sqrt{e}/2)\\sigma$, are universal; if the window mixes growth regimes or the constant relation is wrong, the observed plateaus do not confirm the prediction.","fun_headline_variants_meta":{"raw":{"variants":["Superfast growth: perturbations outpace exponential","Turbulence engine: superfast amplification, saturation","Perturbations grow faster than exponential in turbulence","Superfast amplification: turbulence's self-sustaining secret","Beyond exponential: how turbulence amplifies perturbations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000145,"raw_usage":{"total_tokens":1151,"prompt_tokens":892,"completion_tokens":259,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":508,"completion_tokens_details":{"reasoning_tokens":186}},"tokens_in":508,"tokens_out":259,"duration_ms":3531,"temperature":1.0,"reasoning_tokens":186,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:31:12.595943+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same linear perturbation setup at Reynolds number around 6000 and fit $\\ln(\\delta u(t)/\\delta u(0))/\\sqrt{t}$ over $0.1T_0$ to $0.5T_0$; if the plateau tilts or drifts systematically with the fit window, or if the amplification at fixed $t=0.3T_0$ scales as $Re^{0.38}$ rather than $e^{c\\sqrt{Re}}$, the predicted superfast law is not confirmed.","supporting_citations":[{"cited_title":"Ruelle, Phys","cited_arxiv_id":null,"evidence_quote":"Supplies the baseline exponential prediction with maximal Lyapunov exponent proportional to $\\sqrt{Re}$ that the superfast law extends."},{"cited_title":"Berera and R","cited_arxiv_id":null,"evidence_quote":"Establishes boundedness of the nonlinear perturbation amplification, supporting the distinction between linear and nonlinear growth."},{"cited_title":"Li, Electronic Journal of Differential Equations 2014, 104 (2014)","cited_arxiv_id":null,"evidence_quote":"One of the earlier analytic derivations of the superfast growth law from the Navier-Stokes equations."},{"cited_title":"Li, Nonlinearity 30, 1097 (2017)","cited_arxiv_id":null,"evidence_quote":"Further analytic support for the predicted $e^{\\sigma\\sqrt{Re}\\sqrt{t}+\\sigma_1 t}$ growth law."},{"cited_title":"Inci, Dynamics of Partial Differential Equations 12, 97 (2015)","cited_arxiv_id":null,"evidence_quote":"Rigorous input on the Navier-Stokes dynamics used in the analytic derivation of the prediction."},{"cited_title":"Nowhere-differentiability of the solution map of 2D Euler equations on bounded spatial domain","cited_arxiv_id":"1805.06507","evidence_quote":"Continues the analytic derivation and fixes the constant relation $\\sigma_1=(\\sqrt{e}/2)\\sigma$ used in the prediction."},{"cited_title":"Short term unpredictability of high Reynolds number turbulence --- rough dependence on initial data","cited_arxiv_id":"1702.02993","evidence_quote":"Earlier low-resolution numerical verification that first tested the predicted growth law before the present large-scale simulations."}],"review_version":1}